On Sat, 13 Jan 2024 07:30:29 -0800 Ron / BCLUG via talk <talk@gtalug.org> wrote:
Alvin Starr via talk wrote on 2024-01-12 20:11:
It is defiantly not useful for getting correct technical answers to problems.
agreed. It is accurate when you provide it with code. or actual copy & paste of something you are working with so that it can parse what you are doing wrong or run it's own code on your issue. solving many different types of technical issues it is as useless as search could potentially be and sometimes more so as it could quite convincingly send you on arbitrary solutions that turn out to be completely incorrect.
That's not my experience.
it depends on the technical problem. "AI" does somewhat suck at certain type of problems. Where a coding type "problem" is a shortage of "experience", skill or knowledge type problem, it is quite accurate, in my experience. try asking it about technical problems related to many types of error messages in server logs... try asking it many different types of technical questions where humans generated the issue or where humans are involved in the technical cause for failure.
I guess that depends on the definition of "correct technical answers", because it (i.e. ChatGPT) can be excellent at giving correct answers to technical (coding) problems.
yes, more so accurate if you feed the code (or examples of the code) but - it is not always the most efficient or the "best" solution, but I do agree that it has been a working solution
Got an SQL problem that requires a contrived joining of multiple tables and have tried every left-join / right-join, CTE / subquery combo and just can't get it to work after *hours* of trying? Feed the problem into ChatGPT and get a correct answer in seconds. Got some weird behaviour in JS where asynchronous code and variable scoping issues are colliding to give weird behaviour? Feed the code to ChatGPT, ask "what's wrong", and have it spit out corrected code *AND* and explanation of what's wrong with the code it was provided. Stunning.
good to know, I have not tried feeding it SQL yet, this is actually a good idea and something to remember
And, it's just a generic LLM. I've heard experienced developers saying surprisingly positive things about GitHub's Copilot for quite a while now. As for the SQL issue - all search queries on Qwant / DDG / Google lead to "how to join tables in SQL"; utterly useless. I know that reasonably well. And, who hasn't had a search lead them to StackOverflow where the highest rated answer is strongly condemned further in the comments as being wrong / out of date / insecure, etc.?
Actually, this is an interesting point. Google search seems to prioritise answers from humans and human sources. I searched on Microsoft the other day and was surprised to see that I could supply .js snippets (which I did not code and was too lazy to read through) and receive a correct answer direct from "search" So, us humans will be replaced as 'coders" - Machines will be writing the code which powers machines. Not only is that something for us to understand fully, but we also have to comprehend where we are all choosing to go. It is like watching episodes of "the Traitors" and seeing how the majority votes out a faithful. there is just nothing to do but be along for the ride :)
Lots of incorrect answers supplied by humans.
indeed, if only there was some way to 'sort' or use advanced search to set dates... (to exclude popular answers from 2009) or do more settings on search options... oh, wait.... - and then there are no search results... when is "search" not "search" and just becomes "answer" - interesting! - it is like a mobile phone - it is hardly even a mobile phone any longer, why do so many people still call it a 'phone' or a mobile phone... I think though that I will still be using Google for search, although when looking at it all from my perspective we are all already screwed, unless we can vote out all of the tratitors. (which seems increasingly unlikely)
rb
--- Post to this mailing list talk@gtalug.org Unsubscribe from this mailing list https://gtalug.org/mailman/listinfo/talk